Benchmarked VMESS with transport: QUIC+tls vs WebSocket+tls

[ghost]

Not many people publish benchmarks, so I tried this test.
I'm happy with my current setup - VMESS+WebSocket+NGINX(TLS) - but tried QUIC instead of WebSocket to see whether I can get better speed.
Well, the result is unexpected.

I prepared the following configs, and tried running them using the latest v2ray (v2ray run -c config.json):

• VMESS security=auto & QUIC+tls (I call this "auto-QUIC" below)
• VMESS security=zero & QUIC+tls (I call this "zero-QUIC" below)
• VMESS security=auto & WebSocket+tls (I call this "auto-WS" below)
• VMESS security=zero & WebSocket+tls (I call this "zero-WS" below)

I benchmarked like this:
iperf3 --local--> [v2ray client "dokodemo-door"] --/NSAnet/--> [v2ray server] --local--> iperf3

And here is the benchmark:

• auto-QUIC

    [ ID] Interval           Transfer     Bitrate         Retr
    [  5]   0.00-10.00  sec   600 MBytes   503 Mbits/sec  186            sender
    [  5]   0.00-10.00  sec   596 MBytes   500 Mbits/sec                  receiver

• zero-QUIC

    [ ID] Interval           Transfer     Bitrate         Retr
    [  5]   0.00-10.00  sec   744 MBytes   624 Mbits/sec  139            sender
    [  5]   0.00-10.00  sec   739 MBytes   620 Mbits/sec                  receiver

• auto-WS

    [ ID] Interval           Transfer     Bitrate         Retr
    [  5]   0.00-10.00  sec   928 MBytes   778 Mbits/sec   11            sender
    [  5]   0.00-10.00  sec   921 MBytes   773 Mbits/sec                  receiver

• zero-WS

    [ ID] Interval           Transfer     Bitrate         Retr
    [  5]   0.00-10.00  sec  1.07 GBytes   916 Mbits/sec   17            sender
    [  5]   0.00-10.00  sec  1.06 GBytes   912 Mbits/sec                  receiver

Clearly, WebSocket is faster than QUIC with minimum retry.
What I was expecting is that QUIC would win - I thought UDP is faster than TCP.